1. 1 QoS Schemes for IEEE 802.11 Wireless LAN – An Evaluation by Anders Lindgren, Andreas Almquist and Olov Schelen Presented by Tony Sung, 10 th Feburary 2004

2. 2 Outline  Introduction – Why QoS?  Existing IEEE 802.11 MAC Algorithms DCF and PCF  Proposed QoS Mechanisms Enhanced DCF Blackburst Distributed Fair Queuing  Performance Comparison  Conclusion

3. 3 Introduction – Why QoS?  Shared Medium Medium Utilization can be Low Collision is Possible  To Support Real-time / Multimedia Traffic Require Service Differentiation -> QoS Prioritization Resource Sharing

4. 4 Existing IEEE 802.11 MAC Algorithms  Currently Two Methods are used to provide Medium Access: Distributed Coordination Function (DCF) Mobile Stations try to Compete for Accessing the Medium Point Coordination Function (PCF) Access Point polls the Stations and Grant Access

5. 5 Existing IEEE 802.11 MAC Algorithms  Distributed Coordination Function (DCF) Based on CSMA/CA Algorithm  Sense the Medium before Sending  with Contention Window and Backoff Has Data to Send Sense Medium ( for DIFS ) Has Data to Send Backoff ( for a random time in [0, CW) ) Backoff Timer Suspended DIFS Has Data to Send Backoff DIFS Start TX Timeout Increase CWND & Backoff Start TX ACK Start TX Resume ACK Unknown Delay, Unknown Bandwidth, Low Medium Utilization

6. 6 Existing IEEE 802.11 MAC Algorithms  Point Coordinate Function (PCF) Extends and Coexists with DCF  Controlled by Point Coordinator (i.e. Access Point)  Keeps a List of Stations to be Polled Contention- Free Period Start Backoff DIFS ACK PIFS Poll 1 Station 1: ACK & Data Poll 2 Station 2: ACK & Data Send Beacon Frame Declare End of CFP Contention Free Period (CFP) Higher Utilization, But Delay and Bandwidth may still be an Unknown in High Load situation.

7. 7 Proposed QoS Mechanisms  IEEE 802.11e Enhanced DCF Stations wait for the Channel to become Idle for a pre-defined Time called Inter-frame Spacing (IFS) before sending  Shorter IFS will gain Higher Priority When congested, Backoff time is determined by size of Congestion Window (CW)  Smaller CW will gain Higher Priority

8. 8 Proposed QoS Mechanisms  IEEE 802.11e Enhanced DCF Defines a new IFS called Arbitration IFS Provide Packet Prioritization  Classifies Packets into 8 Different Traffic Classes, Each with different IFS and CW Packet Bursting Has Data to Send Backoff AIFS 1 AIFS 2 Start TX Backoff

9. 9 Proposed QoS Mechanisms  Blackburst Reduce Delay Jitter of High Priority Traffics Send out Black Burst by Jamming the Channel Station that has Waited Longer sends Longer Normal Traffic High Priority Stations Start Black Bursting PIFS Detects! Winner Winner TX

10. 10 Proposed QoS Mechanisms  Distributed Fair Scheduling Prioritization Completely Sacrifice Performance of Low Priority Traffic DFS Provides Proportional Sharing Between Flows according to Assigned Weight Utilizes the Backoff Mechanism of DCF

11. 11 Proposed QoS Mechanisms  Distributed Fair Scheduling Calculate Backoff Interval as follow: Smaller Packets have Higher Chance to be Sent Weight is Added Here Larger Weight means Smaller Backoff Interval, hence Higher Chance for Sending Scale the Backoff Interval to a Reasonable Length Random Variable to Provide Randomness of the Backoff Interval

12. 12 Performance Comparison  Objective Compare  Throughput, Medium Utilization, Collisions, Delay Of  PCF ○, EDCF △, DFS ▓, BB ●  Types of Traffic High Priority (H-P)  300 bytes (Normal Dist.)  25ms Inter-packet Interval (96kb/s) Low Priority (L-P)  800 bytes (Normal Dist.)  50ms Inter-packet Interval (128kb/s)

13. 13 Performance Comparison  Average Throughput All Schemes Achieved Similar Throughput for H-P Traffic BB is best for # H-P Nodes < 13 H-P Traffic Loss Performance 1st in DFS, while maintaining Finite Throughput for L-P Traffic L-P Traffic Starves Rapidly

14. 14 Performance Comparison  Medium Utilization BB has Highest Peak, but Drops at Higher # H-P Nodes ( > 13) EDCF and DFS has Substantially Low Utilization Reasons in the next slides …

15. 15 Performance Comparison  Overhead High when # of H-P Nodes > 13 => Low Utilization

16. 16 Performance Comparison  Collisions EDCF Collides Easily => Low Utilization

17. 17 Performance Comparison  Delay Many H-P Nodes, One CFP cannot Accommodate Medium # of H-P Nodes Many H-P Nodes, Packet Bursting Causes Low Delay for Bursting Packets, and Very High Delay for Waiting Packets Most Cases has Low Delay Worst Case has Delay < 50ms Delay is Proportional to Packet Size, Span Out a Large Range

18. 18 Conclusion  Blackburst and EDCF Starve L-P Traffic Blackburst  Delay is minimal, Best for Real-time  Good at Avoiding Collision EDCF  Starving can be Reduced if using same AIFS for all Traffic in EDCF -> Close to DFS Impl.  Already in IEEE 802.11e  DFS Can be an Alternative if Starving L-P Traffic is Unfavorable  PCF Polling Overhead is High

19. 19 Thank You Questions are Welcomed

20. 20 Appendix

21. 21 Appendix